Intel Gaudi Software Stack Verification

This section provides the required instructions to verify the Intel® Gaudi® software already installed on the system. Start by determining what is already installed on the platform, then decide the next steps for installation. Refer to the Support Matrix for the list of supported and validated FW and SW components.

If you are just interested in running Intel Gaudi’s Docker images, refer to Pull Prebuilt Containers.

Note

The versions listed in the Support Matrix are the validated configurations from Intel Gaudi. Using configurations not tested by Intel Gaudi may not be functional or performant.

Check Existing Software Installation

This section shows how to check the existing SW and FW components on the platform. In general, it is advised for a SW developer to use the Intel Gaudi Docker image to ensure Intel Gaudi SW and PyTorch framework are installed.

To verify the current Intel Gaudi Software version, run the hl-smi tool. You will need to use the correct software version to run the matching installer or Docker image. Use the HL-SMI Version at the top of the output. For example, if the installed version is 1.15.0, you should see the below:

   HL-SMI Version:       hl-1.15.0-XXXXXXX
   Driver Version:       1.15.0-XXXXXX

To verify the system software, run the following command:

apt list --installed | grep habana

If the installation is successful, you should see the habanalabs-dkms driver and the following software components at the root system. The habanalabs-rdma-core package is installed on Gaudi 2 only.

 habanalabs-container-runtime/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-dkms/focal,focal,now 1.15.0-479 all [installed]
 habanalabs-firmware/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-firmware-odm/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-firmware-tools/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-graph/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-rdma-core/now 1.15.0-479 all [installed]+
 habanalabs-qual/focal,now 1.15.0-479 amd64 [installed]
 habanalabs-thunk/focal,focal,now 1.15.0-479 all [installed]

Note

In the Intel Gaudi Docker image, only the habanalabs-firmware-tools, habanalabs-graph, habanalabs-rdma-core, and habanalabs-thunk will be visible.

To verify the PyTorch framework used on the system, run the following command:

     pip list | grep habana

     habana-media-loader     1.15.0-479
     habana-pyhlml           1.15.0-479
     habana-torch-dataloader 1.15.0-479
     habana-torch-plugin     1.15.0-479
     lightning-habana        1.3.0

Next Steps

After verifying the installed versions, you can do the following: